Contracting for LLM Delegation: Moral Hazard in Technology and Effort Choice
3.40T1 sourcearXiv cs.MA
Source record
Published by arXiv cs.MA (T1 source). The original is at https://arxiv.org/abs/2608.18232.
Pipeline notes
The summary and note below are generated by the signal pipeline — they are Beyond Desk’s reading, not quotations from the source.
SummaryThe paper extends the Principal-Agent framework to LLM delegation, modeling agents that choose both a model and an effort level (e.g., token budget). It derives optimal linear contracts characterized by a threshold reward share triggering technology switching, then validates the theory using bandit algorithms on MATH and MMLUPro benchmarks with open-weight LLMs.
Why it mattersFormal economic lens on how to structure incentives around LLM choice and effort, grounded in both theory and empirical benchmark calibration. Useful framing for anyone designing or evaluating agent delegation contracts.
Cited by
No citations on record.